Reconsidering Degeneration of Token Embeddings with Definitions for Encoder-based Pre-trained Language Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zhang, Ying, Li, Dongyuan, Okumura, Manabu
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913548628656128
author Zhang, Ying
Li, Dongyuan
Okumura, Manabu
author_facet Zhang, Ying
Li, Dongyuan
Okumura, Manabu
contents Learning token embeddings based on token co-occurrence statistics has proven effective for both pre-training and fine-tuning in natural language processing. However, recent studies have pointed out that the distribution of learned embeddings degenerates into anisotropy (i.e., non-uniform distribution), and even pre-trained language models (PLMs) suffer from a loss of semantics-related information in embeddings for low-frequency tokens. This study first analyzes the fine-tuning dynamics of encoder-based PLMs and demonstrates their robustness against degeneration. On the basis of this analysis, we propose DefinitionEMB, a method that utilizes definitions to re-construct isotropically distributed and semantics-related token embeddings for encoder-based PLMs while maintaining original robustness during fine-tuning. Our experiments demonstrate the effectiveness of leveraging definitions from Wiktionary to re-construct such embeddings for two encoder-based PLMs: RoBERTa-base and BART-large. Furthermore, the re-constructed embeddings for low-frequency tokens improve the performance of these models across various GLUE and four text summarization datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2408_01308
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Reconsidering Degeneration of Token Embeddings with Definitions for Encoder-based Pre-trained Language Models
Zhang, Ying
Li, Dongyuan
Okumura, Manabu
Computation and Language
Learning token embeddings based on token co-occurrence statistics has proven effective for both pre-training and fine-tuning in natural language processing. However, recent studies have pointed out that the distribution of learned embeddings degenerates into anisotropy (i.e., non-uniform distribution), and even pre-trained language models (PLMs) suffer from a loss of semantics-related information in embeddings for low-frequency tokens. This study first analyzes the fine-tuning dynamics of encoder-based PLMs and demonstrates their robustness against degeneration. On the basis of this analysis, we propose DefinitionEMB, a method that utilizes definitions to re-construct isotropically distributed and semantics-related token embeddings for encoder-based PLMs while maintaining original robustness during fine-tuning. Our experiments demonstrate the effectiveness of leveraging definitions from Wiktionary to re-construct such embeddings for two encoder-based PLMs: RoBERTa-base and BART-large. Furthermore, the re-constructed embeddings for low-frequency tokens improve the performance of these models across various GLUE and four text summarization datasets.
title Reconsidering Degeneration of Token Embeddings with Definitions for Encoder-based Pre-trained Language Models
topic Computation and Language
url https://arxiv.org/abs/2408.01308